Methods 2: study populations

Identification of women who used a pool during labour

Women who used any form of water immersion during labour were pragmatically defined as any women for whom water immersion analgesia, or a waterbirth, had been recorded in the maternity record.

Identification of women who gave birth in water

Women were pragmatically defined as having given birth in water if this was recorded in their maternity record. To capture births commenced in, but completed, out of water, such as in the event of shoulder dystocia or previously unrecognised breech presentation, ‘waterbirth’ was defined in the study as ‘a birth in which the fetus is partially or totally expelled under water’. A field recording ‘partial births’ in water was only available from records after site opening. Partial births in water occurring prior to site opening were identified from text entries in the records.

Identification of women with complexities

The criteria of ‘low risk’ within maternity care is one of exclusion. The NICE Intrapartum Care Guidelines39 list conditions that, if present, should be regarded as an indication to either advise birth in an OU (see Appendix 5, Tables 30 and 31), or that suggest individual assessment should be undertaken prior to making a recommendation on the planned place of birth (see Appendix 5, Tables 32 and 33).

To identify women with any of the conditions listed by NICE indicating recommended birth in an OU, or individual care planning, fields held in the ‘medical and surgical history’ section, and other antenatal sections of records completed during the pregnancy were mapped to the NICE guidelines. For births after site opening, midwives also completed the new fields providing information on complexities present at the time of pool entry.

In combination, these fields were used to identify women with an identified complexity who used a pool in labour.

It was anticipated that the classification of risk may have differed between these two sources.

Risk classification categorisation based on the existing MIS antenatal fields is likely to provide a lower threshold for risk, potentially identifying women who experienced a complication in the past or during pregnancy, but for whom this was no longer present at pool entry, for example, a woman with an episode of hypertension during pregnancy, but who was later normotensive. Risk classification categorisation used by the midwife at pool entry is likely to provide a more pragmatic definition reflecting the opinion of the midwife providing intrapartum care.

For women who gave birth prior to site opening, for whom complexities relating to the time of pool entry were not available, if there was any record of a relevant factor in the antenatal notes, the woman was classified as complex.

For women who gave birth after site opening, if factors that cannot change over time were recorded in the antenatal notes, the woman was classified as complex regardless of whether this was also identified by the midwife providing intrapartum care, for example, a previous caesarean section.

For women who gave birth after site opening and where factors that can change over time were recorded in the antenatal notes, when these were not identified by the midwife providing intrapartum care, the woman was classified as without complexity, for example, hypertension, suspected macrosomia.

Medical and surgical information storage within Euroking

Magentus (formerly WS) notified users of the MIS, in a service bulletin dated 2 July 2023, of issues within their system in the management of pregnancy-specific data which, when carried over to the next pregnancy, overwrite the data relating to past pregnancies. These data included the data held within the ‘Medical & Surgical History’ menu. It was subsequently confirmed that all Euroking MISs, across all dates, were similarly affected.

Implications for the POOL data set

For convenience, when a woman ‘books’ for a second or subsequent pregnancy, the medical and surgical history section is opened, showing the entries from the previous pregnancy. This saves time and duplication of entry, as only required changes need to be entered. As the medical and surgical history had been stored as a single record at patient level, the original medical and surgical history, as recorded in the previous pregnancy, was overwritten.

By July 2023, when the study team became aware of how the data in the medical and surgical menu were held, all study data had been extracted and received by the study team and these data were used to risk classify women.

The fields contained in the ‘Medical & Surgical History’ were:

• cardiac problem
• haematological problem
• respiratory problem
• hepatic problems
• renal problems
• endocrine problem
• neurological problem
• musculoskeletal problem
• gynaecological problem
• operations
• infections
• mental health problem
• thromboembolic disorder
• genetic disorder
• thyroid problems
• deep-vein thrombosis risk assessment.

The POOL Study spans births occurring between 1 January 2015 and 30 June 2022. Many of the cohorts will have had more than one pregnancy during this period. As each woman is allocated with a different ID for each pregnancy, and due to the anonymised nature of the data, women with more than one pregnancy could not identified.

As a result, it was not possible to determine whether the medical and surgical history section contained within each maternity record received was correct or related to a future pregnancy for the same woman. For example, if a woman had three pregnancies and births within the POOL Study data set and was detected to carry group B streptococcus (GBS) only on her third pregnancy, as the medical and surgical data were held at patient level, within the POOL data set, the woman would be identified as a GBS carrier for all three pregnancies.

Impact on analysis relating to women without risk factors who used a pool in labour

The impact of the overwriting of medical and surgical information in the MIS was that factors recorded or developing during subsequent pregnancies were included in the history of all past pregnancies. As all women with factors identified in their records were excluded from the primary analysis, the impact of this was that more women than were necessary were excluded from the without complexity groups.

It is also possible that a medical problem recorded as present during one pregnancy will have been removed from the record during a future pregnancy. However, as the history taken at booking captures the ‘current or past medical and surgical history’, conditions would normally be expected to remain in the record even if those are no longer an active concern.

Impact on analysis relating to women with risk factors who used a pool in labour

One of The POOL Study objectives was to establish the characteristics of, and outcomes for, women with identified complexities at labour onset, who used a pool during labour. Complexities recorded outside of the medical and surgical history menu in the antenatal record, and in the new field relating to the time of pool entry, were not compromised. All data held within the medical and surgical history fields were considered to be potentially compromised.

The impact was that women who had been classified as having complexities solely based on data in the potentially compromised fields needed to be removed from this analysis. This objective was answered using the prospective data only.

Impact on analysis relating to women who did not use a pool in labour

One of the POOL Study objectives was to describe the proportion and characteristics of women who used a pool for labour or birth compared to women who do not use a pool. It had been planned to identify women without risk factors who had not used a pool in labour, but this was not undertaken.

Study population groups

Women were separated into groups, depending on (1) their pool use for labour and/or birth; (2) the presence of complexities in antenatal records; (3) whether transfer to obstetric care for additional monitoring or interventions occurred prior to birth and (4) among women not transferred to obstetric care prior to birth, whether any midwife concerns were present before birth (Figure 3 and Table 2). Records were categorised into the following groups:

FIGURE 3 Study population: the five study groups. AMU, alongside midwifery unit; FMU, freestanding midwifery unit; OU, obstetric unit. (Figure not reproduced here; see the original article.)

• Group 1 – without antenatal complexity – waterbirths

This comprised women who used a pool in labour, with no risk factors in their antenatal record or recorded at pool entry, without additional monitoring or interventions before birth, who had a waterbirth.

• 1a – no clinical concern recorded by the midwife as being present prior to birth.
• 1b – clinical concern recorded by the midwife as being present prior to birth.

• Group 2 – without antenatal complexity – births out of water

This comprised women who used a pool in labour, with no risk factors in their antenatal record or recorded at pool entry, without additional monitoring or interventions before birth, who gave birth out of water.

• 2a – no clinical concern recorded by the midwife as being present prior to birth.
• 2b – clinical concern recorded by the midwife as being present prior to birth.

• Group 3 –without antenatal complexity – with additional monitoring or interventions before birth

This comprised women who used a pool in labour, without risk factors in their antenatal record or recorded at pool entry, with additional monitoring or interventions before birth, including women who were transferred to obstetric care during labour.

• Group 4 – with antenatal complexity

This comprised women who used a pool in labour with recorded risk factors in their antenatal record or recorded at pool entry.

• Group 5 – women who did not use a pool during labour or birth, without antenatal complexity and with complexity.

Identification of women in group 1

Group 1 included women who used a pool in labour, with no risk factors in their antenatal record or recorded at pool entry, without additional monitoring or interventions before birth, who had a waterbirth.

Women were allocated to group 1 if all the following points applied:

• an absence of risk factors in their antenatal record
• an absence of risk factors at pool entry (for births after site opening)
• record of pool use in labour
• record of a waterbirth
• an absence of any record of an intervention which would move the woman to group 3.

Women allocated to group 1 remained in group 1 if any of the following were recorded:

• non-significant meconium (also described as light, or grade 1 or not described)
• women with recorded delay in the first stage of labour (but without the use of augmentaion with oxytocin)
• CTG monitoring performed with normal findings and discontinued
• significant meconium (or similar description) only seen at, or after, delivery of the fetal head
• partial births in water with breech presentation or shoulder dystocia
• tight nuchal cord
• any third-stage complications, for example PPH.

Women were identified by using a combination of the waterbirth field in maternity records (for all records); and, to capture births commenced in, but completed, out of water (such as in the event of shoulder dystocia or previously unrecognised breech presentation), the new waterbirth field was used: ‘A birth in which the fetus is partially or totally expelled under water’. This information was only available from records after site opening. For the period of data collection where these additional data was not available, we took any recording of waterbirth as such. A small number of additional partial waterbirths were picked up in the additional field ‘POOLLeftPoolNoReturn’, where there was clear evidence of a partial waterbirth recorded in the additional text (e.g. slow delivery of shoulders, stood up in pool to deliver and left pool to assist with delivering body).

Identification of women in group 2

Group 2 included women who used a pool in labour, without complexities recorded in their antenatal record or recorded at pool entry, without transfer to obstetric care during labour, who gave birth out of water.

Women were allocated to group 2 if all the following points applied:

• An absence of risk factors in their antenatal record.
• An absence of risk factors at pool entry (for births after site opening).
• Record of pool use in labour.
• No record of a waterbirth
• For women giving birth after site opening, the ‘Reason for leaving pool before birth’ is recoded as:
  1. left pool for vaginal examination/use bathroom and did not return’
  2. maternal decision to leave pool and did not return
  3. left pool for further analgesia and did not return
  4. planned to labour but not give birth in water.
• An absence of any record of an intervention which would move the woman to group 3.

Women allocated to group 2 remained in that group if any of the following points were recorded:

• Non-significant meconium (also described as light, or grade 1 or not described).
• Women with recorded delay in the first stage of labour (but without augmentaion with oxytocin).
• CTG monitoring performed, but it was normal and discontinued
• Significant meconium (or similar description) only seen at, or after, delivery of the fetal head.
• Partial births in water with breech presentation or shoulder dystocia.
• Tight nuchal cord.
• Any third-stage complications, for example PPH.

Identification of women in groups 1a and 2a

Groups 1a and 2a were subgroups of groups 1 and 2. Women in groups 1 and 2 were eligible to be in group 1a or 2a, respectively, if there were no recorded clinical concerns prior to birth.

Identification of women in groups 1b and 2b

Groups 1b and 2b were subgroups of groups 1 and 2. Women in groups 1 and 2 were eligible to be in group 1b or 2b, respectively, if midwives recorded that there had been some midwife concern occurring before or during some births, but these either did not require, or had no time for, transfer to obstetric care.

To identify midwife concern present before birth, but that did not result in recorded obstetric involvement in a woman’s care, we used a combination of existing fields (e.g. maternal/fetal intrapartum problems: ‘Were there any maternal/fetal problems during labour?’) and a new field (POOLLabourComplications), both with an option for open text comments.

Groups 1b and 2b included women where the midwife recorded that, prior to birth, there was:

• an abnormal fetal heart rate (HR) (fetal compromise, bradycardia, decelerations, fetal tachycardia, irregular heartbeat, rise in fetal heart baseline and fetal distress)
• some delay in second-stage progress (without instrumental birth)
• blood-stained liquor
• significant meconium-stained liquor (thick) seen prior to delivery of the fetal head
• intrapartum haemorrhage
• maternal hypertension/tachycardia/pyrexia/suspected infection (without treatment)
• reduced fetal movements
• other concern.

Identification of women in group 3

Women who used a pool during labour, without complexities identified in the antenatal records/or by the midwives at the time of pool entry, and who received additional monitoring or obstetric interventions before or during birth formed group 3.

Women were allocated to group 3 if all the following points applied:

• an absence of risk factors in their antenatal record
• an absence of risk factors at pool entry (for births after site opening)
• record of pool use in labour
• record of any of the following:
  • pain relief incompatible with use in water (e.g. epidural, remifentanil and pudendal block) before birth
  • Oxytocin augmentation of labour
  • CTG described as pathological, suspicious, abnormal or non-reassuring
  • breech presentation identified during labour with obstetric involvement in care
  • caesarean section or instrumental birth
  • transfer from home or a MLU during first or second stage of labour.
  • maternal pyrexia with investigation or treatment
  • receipt of instrumental variable (IV) fluids
  • antibiotics for suspected chorioamnionitis/sepsis
  • placental abruption
  • intrapartum eclampsia
  • additional obstetric monitoring or treatment.

Identification of women in group 4

Women who used a pool in labour, who had recorded risk factors in their antenatal record or (for births after site opening) recorded at pool entry, formed group 4.

Identification of women in group 5

Women with no record of pool use during labour or birth were allocated to group 5.

Study populations for primary analysis

Two primary analyses were undertaken, as it was considered to be important to report ‘real life’, where concerns may be present prior to birth, and also to report on any impact of giving birth in water in the absence of any midwife concerns (Table 3).

Pooling of investigational sites

Records from all sites were pooled for the analysis. Sites were identified by their site ID number (and not named) and were included in the regression models as a random factor.

Study opt-out

Data held in a site’s MIS relating to women who had opted out of the study were not extracted. The number of women who opted out per site over the study period are reported. Where sites changed their MIS supplier during the period of the study, births recorded in the new MIS were also not extracted.

Statistical analysis

Prior to commencement of the analysis, a statistical analysis plan (SAP) was produced (https://osf.io/pqadm), reviewed by the study management group (SMG), signed off by the chief investigator and statistician and approved by the Study Steering Committee (SSC) before data collection was completed and examined.

Sample size

The non-inferiority of birth in water compared to birth on land on rates of OASI was examined by parity. The Birthplace in England study10 found that 4.6% and 1.6% of ‘low-risk’ nulliparous and parous women in spontaneous labour, respectively, sustained OASI. A sample size of 15,000 nulliparous and 15,000 parous women (7500 each water and land) without antenatal complexities, and who did not require additional monitoring or interventions before birth, was required to obtain 90% power and a one-sided 95% CI around a treatment difference of zero. Non-inferiority margins of ≤ 1% (OR ≤ 1.23) and ≤ 0.6% (OR ≤ 1.38) were taken as clinically non-significant among nulliparous and parous women without antenatal complexities, respectively. Since nulliparous women birthing in water are regarded as the least prevalent of the four groups, a data collection period providing data on 7500 nulliparous women giving birth in water ensured adequate numbers in the other three, more prevalent groups. These data were combined to assess the effects averaged across both strata at an increased power, with a combined required sample size of 30,000 women without complexities. We have assumed that 25% of the 6600 waterbirths recorded in Euroking in 2015 were nulliparous women (1650/annum). Allowing for staggered site set-up, 6 years of combined retrospective and prospective data collection was required (January 2015–June 2022). With potential study sites, collectively undertaking 6037 waterbirths in 2016, it was believed that the study would have sufficient power to answer this important clinical question.

Based on estimates from the Birthplace in England study for the infant primary outcome, an estimate of 5% was used for the proportion of infants born to low-risk mothers experiencing ‘adverse infant outcome or treatment’. A non-inferiority margin of ≤ 1.0% (OR ≤ 1.21) was taken as clinically non-significant. A sample size of 16,200 infants (8100 per group water/land) was required to have 90% power and a one-sided 95% CI around a treatment difference of zero.

Descriptive analysis

The numbers of records received from all sites, as well as the total number of women and babies for groups 1–5, were described and depicted in a flow chart. We have described opt-outs by each NHS site alongside the number and rate of women not using a pool (group 5), using a pool (groups 1–4), by risk status (low risk, groups 1–3 and underlying condition, group 4) and waterbirth (group 1). Maternal and infant characteristics, such as age, parity and ethnicity of all women giving birth in the study sites during data collection, were summarised by study population group (Table 4).

Primary analysis

The primary analyses were based on a non-inferiority test of births occurring in water (group 1) versus births occurring out of water (group 2) (reflecting real life practice) comparing:

1. the proportion of mothers who have OASI (based on retrospective and prospective maternity data)
2. the proportion of infants with a composite outcome of ‘adverse infant outcome or treatment’ (based on prospective maternity and NNRD-provided data).

Non-inferiority testing explored whether birth in water was not worse than birth out of water by more than the predefined non-inferiority margin and established at the 5% (one-sided) of the upper limit of the CI for the difference between groups.

• Maternal outcome: Non-inferiority would be concluded if the upper limit of the 95% CI for the difference in the proportion of OASI between the groups is < 1.0% (OR ≤ 1.23) in nulliparous low-risk women and < 0.6% (OR ≤ 1.38) in parous women. The data were then combined to assess the effects averaged across both strata.
• Infant outcome: Non-inferiority would be concluded if the upper limit of the 95% CI for the difference in infant outcome between the groups is < 1.0% (OR ≤ 1.21).

To test the primary hypothesis of non-inferiority between babies born in water versus leaving the pool before birth, both the maternal and infant primary outcomes were evaluated for non-inferiority using logistic regression models. All analyses used a mixed-effects two-level regression model to allow for clustering of outcomes by site. We did not account for the clustering of infants within mothers within site, as we anticipated a small number of multiple births.

Three sets of ORs were presented: unadjusted OR (UOR), adjusted OR (aOR) for selected confounders (no imputation) and aOR for selected confounders (with imputation). ORs were presented alongside a one-sided 95% CI (−∞ to upper limit), where the upper limit was compared to the predefined non-inferiority margins.41 Confounders of interest were year of birth, ethnic group, maternal age, deprivation quintile, parity, gestational age, BMI, birthweight and midwife concern present at birth (post hoc adjustment).

Unit of analysis

For maternal outcomes, women were the unit of analysis (denominator), and those who left the pool to give birth were used as the reference group for all comparative analyses. For infant outcomes, babies were the unit of analysis (denominator), and those with mothers who left the pool to birth were used as the reference group for all comparative analyses.

Secondary analyses

If non-inferiority was demonstrated, then a superiority analysis was conducted as a secondary analysis of the primary outcomes, again using logistic regression and presented as UORs and aORs, alongside a two-sided 95% CI.

Further secondary analyses of components of the infant primary composite outcome were undertaken using both retrospective and prospective maternity and NNRD data for the following outcomes: NNU admissions requiring respiratory support and intrapartum stillbirth or early neonatal death, as these were captured over both periods of data collection in both sources. This was done to increase the sample size and the power of these analyses. However, this outcome did not include administration of intravenous antibiotic within 48 hours of birth among babies not admitted to a NNU, nor babies that died without admission to a NNU.

Subgroup analysis

Parity

A planned and powered subgroup of the primary maternal outcome was conducted to compare rates of OASI separately for primiparous and multiparous women. The relationship between the proportion of women using a pool during labour at individual sites and the incidence of adverse maternal and infant primary outcomes were described and explored. A planned subgroup analysis of the primary infant composite outcome was also conducted to compare rates separately for infants born to primiparous and multiparous women. These preplanned analyses were conducted by the inclusion of appropriate interaction terms (waterbirth exposure × parity) in the regression models. Results were presented using interaction coefficients, 95% CI and p-value.

Delivery of placenta in water

To explore potential risks of delivery of the placenta in water, an important subgroup was that of women who birthed in water, by whether the placenta was delivered in water, or the woman left the pool during the third stage. We examined the primary maternal and infant outcomes and PPH of ≥ 1000 ml between these two groups.

Sensitivity analyses

For both maternal and infant primary outcomes, several sensitivity analyses were performed to assess the robustness of the results to factors which may introduce bias (i.e. definition of risk and the study populations, maternal characteristics associated with waterbirth and fetal HR concerns).

Clinical need

The study has reported outcomes reflecting clinical practice, including where a risk was described as having been present at the time of pool entry as its primary analysis. It also reports outcomes among women without any risk factors in either antenatal records or at pool entry as a sensitivity analysis among group 1a (birth in pool + no midwife concerns) versus group 2a (birth in/out of pool + no midwife concerns).

Propensity score analysis

In addition to adjusting for maternal and infant characteristics for each outcome as in our primary analysis approach, we reweighted on the propensity score using inverse probability weighting (IPW) so that no controls were excluded. Incorporating propensity scores, that is the ‘propensity’ of a woman to choose a waterbirth, in the analysis is a way of controlling for this bias and can produce estimates with greater precision. Propensity scores were derived from logistic regression models, with waterbirth as the outcome and this included all maternal predictors of waterbirth.

Instrumental variable analysis

The proportion of women using water for labour or birth at each unit was used as an IV in this analysis, and the assumptions were tested that the IV variable was (1) associated with the exposure (i.e. remaining in water for birth) and (2) not associated with the outcome.

Missing data

Empty cells were distinguished by:

1. sites not collecting certain fields (partial or full study period) or entirely halting data collection (e.g. ceasing to use the Euroking MIS)
2. cells that were expected but were empty (coded as NULL).

For (2), fields were separated into those:

1. Expected to be well completed (e.g. mode of birth, birthweight and breastfeeding). Empty cells were defined as truly missing and imputation considered (see Case study findings).
2. Likely to only be completed when an event has occurred (e.g. hypertension). Empty cells were defined as an absence of event
3. Only expected when relevant prescreening questions are used (e.g. duration of antibiotics only applicable for those that receive antibiotics). Empty cells were defined as ‘not applicable’ unless the screening question was positive in which case an empty field would be defined as missing.

Where appropriate, we used multiple imputation methods [using the mi command in Stata^® (StataCorp LP, College Station, TX, USA)], assuming that data were are missing at random or were missing completely at random. To assess the effect of missing data on the results of the primary analysis, a sensitivity analysis was planned using multiple imputation techniques to impute missing data for each of the potential confounders included in the adjusted regression models under the assumption that the data were missing at random. This assumes that the reason data are missing is not dependent on the value of the missing data if it were known. Missing outcome data would not be imputed since no information is gained, since the standard errors from imputation are likely to be larger than those of the complete case analysis.

Multiple imputation with chained equation42 was used to impute values for the missing data (10 imputations) under the missing at random assumption, with parameter estimates and their standard errors combined using Rubin’s rules.

Secondary outcomes

Secondary outcomes underwent non-inferiority testing as for the primary outcomes, and these are listed in Appendix 6 alongside the study population used and analysis approach taken. In brief, the method of analysis is dependent on the outcome type, for example, binary (yes/no, presence or absence of events), continuous, ordinal and count data. Binary outcomes were modelled using logistic regression models and effect estimates, presented as ORs comparing the odds of an event in waterbirth compared to land. For continuous outcomes, multilevel linear models were fitted, and the results were presented as a difference in means (waterbirth minus birth on land). Ordered categorical data, such as perineal trauma, were modelled using ordinal regression. Count data were analysed using a Poisson multilevel model. If the distribution of events displayed signs of overdispersion (greater variance than might be expected in a Poisson distribution), then a negative binomial model (NBM) was used; estimates were presented as the incidence rate ratio (IRR) in waterbirth compared to on land. All parameter estimates were accompanied by a one-sided 95% CI and p-value.

The rate of PPH was compared for women who had their placenta delivered in water and those leaving the pool during the third stage, and results were presented as an aOR accompanied by a two-sided 95% CI.

Bias

There is a potential for reporting bias of the risk categorisation at pool entry collected by midwives after site opening, as this will usually be recorded after the outcome of the baby/mother was known. To examine this bias, we planned to examine trends in the incidence of overall risk and by categories over the study period and by the data sources (Euroking existing fields and midwives’ entry) to detect any increases caused by ‘diagnostic drift’. Due to the issue with the risk classification categorisation in Euroking, this analysis could not be undertaken.

Exploratory analyses

Several exploratory analyses were planned and prespecified in the SAP:

• Examination of the impact of the COVID-19 pandemic on the rates of pool use, waterbirths and other procedures that might have been altered by the pandemic (such as the rate of women under obstetric care, inductions, use of oxytocin augmenation of labour, etc.) by examining trends overall and by sites.
• Examination of the trends in antibiotic use in neonates and whether change in trends can be attributed to a change in use of guidelines (from NICE to sepsis calculator). We also examined the association with the factors associated with rates of antibiotic use.
• Examination of the factors associated with rates of pool use in individual maternity units.
• Characterisation of sites and mothers for pool use in low-risk women.
• The women with risk factors recorded in the maternity notes (group 4) who used a pool during labour were characterised by whether they gave birth in the pool or not.

Software

Data analysis was conducted in IBM SPSS Statistics version 26 (IBM Corporation, Armonk, NY, USA) and Stata version 16.

Reporting guidelines

The reporting of The POOL Study follows the REporting of studies Conducted using Observational Routinely collected health Data (RECORD) guidelines (see Appendix 7) and also Consolidated Standards of Reporting Trials (CONSORT) guidelines on reporting for equivalence and non-inferiority trials.43

Changes to statistical methods from the protocol and statistical analysis plan

A summary to the changes to the outcomes and approach to analyses are given in Appendix 8, and modifications made by the approach to analyses are given in Appendix 6.